Analysis Ready Data (ARD) and CEOS CARD4L Quality Standards
Eliminate custom preprocessing pipelines by verifying CEOS CARD4L compliance, sub-pixel co-registration, radiometric calibration, and QA bitmasks.
What Makes Satellite Imagery Truly Analysis Ready (ARD)
Raw Level-1 scenes demand orthorectification, radiometric calibration, atmospheric correction, and cloud masking before time-series analysis. Analysis Ready Data (ARD) automates these steps upstream so pixels enter GIS and ML pipelines immediately.
CEOS CARD4L establishes auditable Threshold and Target specifications for Optical Surface Reflectance, Surface Temperature, Normalized Radar Backscatter (NRB), and Polarimetric Radar.
- Threshold sets mandatory geometric, radiometric, and metadata baselines.
- Target adds stricter BRDF normalization and per-pixel uncertainty.
- Standardized QA layers flag clouds, cloud shadows, terrain shadows, and water.
Geometric Co-Registration and Circular Error 90 (CE90)
Inter-date spatial shifts generate false change alerts along roads and field edges. Horizontal accuracy is quoted as Circular Error at 90% confidence (CE90), while change detection demands sub-pixel co-registration (<= 0.5 pixel RMSE) to a common reference grid.
- Confirm ortho-DEM resolution matches or exceeds sensor GSD.
- Keep map projections (e.g., UTM WGS84) uniform across stacks.
- Verify CE90 tests include terrain relief checkpoints.
Radiometric Consistency and BRDF Normalization
Off-nadir viewing and seasonal sun angles distort reflectance via surface BRDF anisotropy. Advanced optical ARD applies BRDF normalization, while SAR CARD4L-NRB applies terrain flattening so sloped fields match flat terrain.
Auditing Lineage and Per-Pixel Uncertainty Metadata
CARD4L embeds DEM version, atmospheric model, and orbit lineage inside STAC or ISO 19115 records so insurance, carbon, and regulatory audits stay reproducible.
CEOS CARD4L Product Family Requirements Summary
| CARD4L Product Family | Core Physical Measurement | Mandatory Corrections | Key Per-Pixel Quality Masks |
|---|---|---|---|
| Surface Reflectance (SR) | Bottom-of-atmosphere reflectance | Ortho + TOA + atmospheric correction | Cloud, cloud shadow, snow, cirrus, saturation |
| Surface Temperature (ST) | Land surface temperature (K) | Emissivity + water vapor correction | Cloud, adjacency, emissivity uncertainty |
| Normalized Radar Backscatter (NRB) | Terrain-flattened gamma0 | Precise orbit + RTC + calibration | Layover, radar shadow, local incidence angle |
| Polarimetric Radar (POL) | Covariance / coherency matrix | Polarimetric calib + RTC geocoding | Layover, shadow, effective looks |
| Geocoded SLC (GSLC) | Complex phase on map grid | Precise orbit + terrain phase | Layover, shadow, phase validity |
CEOS specifies Threshold (baseline) and Target (advanced BRDF/uncertainty) compliance tiers per product family.
Worked Illustration: Evaluating Horizontal CE90 and Co-Registration Tolerance
A 3.0 m ARD stack checked against 20 GCPs yields RMSE_x = 0.85 m and RMSE_y = 0.95 m. We compute radial RMSEr, CE90, and pixel tolerance.
- Compute radial RMSE: sqrt(0.85^2 + 0.95^2) = sqrt(1.625) = 1.275 m.
- Convert to CE90 (NSSDA factor 1.5175): 1.5175 x 1.275 m = 1.93 m.
- Express error in 3.0 m pixel units: 1.275 m / 3.0 m = 0.425 pixels.
- Verify against 0.50-pixel threshold: 0.425 <= 0.50 confirms sub-pixel co-registration.
Calculated Result & Takeaway: The stack achieves 1.28 m RMSEr (1.93 m CE90), equal to 0.43 pixels—meeting sub-pixel alignment for change detection.
ARD & CARD4L Procurement Verification Checklist
- Confirm whether vendor ARD is CEOS CARD4L self-assessed or peer-reviewed.
- Verify delivery includes per-pixel QA cloud, shadow, and layover masks.
- Check CE90 and inter-scene co-registration RMSE in pixel units.
- Record atmospheric model and calibration baseline in pipeline logs.
- Confirm ortho-DEM source used for terrain correction.
- Ensure nodata and scale factors are explicit in STAC metadata.
Frequently Asked Questions
Are all vendor 'ARD' products equivalent to CEOS CARD4L?
No. Checking CEOS CARD4L compliance ensures standardized corrections, lineage metadata, and per-pixel QA masks.
Why does relative co-registration matter more than absolute CE90?
Uniform 2 m shifts keep time-series pixels aligned, whereas random shifts between dates trigger false change edges everywhere.
Is Landsat Collection 2 Level-2 CEOS CARD4L compliant?
Yes. USGS Landsat Collection 2 Level-2 Surface Reflectance and Temperature products are peer-reviewed CARD4L Threshold compliant.
Do I still need cloud masking on CARD4L imagery?
Yes. CARD4L provides standardized cloud and shadow QA bitmasks so code can filter bad pixels automatically.
Primary References & Standards
- CEOS Analysis Ready Data (CEOS-ARD) Product Family Specifications.
- USGS Landsat Collection 2 Level-2 CEOS CARD4L Assessment.
- ASPRS Positional Accuracy Standards for Digital Geospatial Data.